From 7b32ed69f7dfdc4b2a4cb905f048acdb812ce2b4 Mon Sep 17 00:00:00 2001 From: Dave Date: Wed, 25 Feb 2026 12:02:35 +0000 Subject: [PATCH] story-kit: create 180_bug_web_ui_permissions_handling_unreliable --- ..._web_ui_permissions_handling_unreliable.md | 35 +++++++++++++++++++ 1 file changed, 35 insertions(+) create mode 100644 .story_kit/work/1_upcoming/180_bug_web_ui_permissions_handling_unreliable.md diff --git a/.story_kit/work/1_upcoming/180_bug_web_ui_permissions_handling_unreliable.md b/.story_kit/work/1_upcoming/180_bug_web_ui_permissions_handling_unreliable.md new file mode 100644 index 0000000..8df43d3 --- /dev/null +++ b/.story_kit/work/1_upcoming/180_bug_web_ui_permissions_handling_unreliable.md @@ -0,0 +1,35 @@ +--- +name: "Web UI permissions handling unreliable" +--- + +# Bug 180: Web UI permissions handling unreliable + +## Description + +Permissions handling in the web UI chat has multiple issues that make it unreliable. This is a tracking bug to collect specific problems as they're encountered. + +Known issues so far: +1. **Permission dialog not appearing**: When the agent needs tool approval, the permission request dialog sometimes doesn't pop up in the web UI, leaving the agent blocked with no way for the user to approve/deny. This was the main problem over the past week. +2. **Batch tool calls failing**: When the agent fires many parallel Bash calls (e.g. 11 at once), the permission/hook validation system fails with cryptic errors (`invalid_union`, `Invalid input: expected "allow"`), cascading to cancel all sibling calls. Individual calls with the same commands work fine. Unclear if this is a Claude Code SDK limitation, a hook validation issue, or a web UI problem. + +## How to Reproduce + +Issue 1: Start a chat session using claude-code provider, trigger a tool call that requires permission (e.g. a Bash command not in the allow list). Observe that the permission dialog sometimes fails to appear. + +Issue 2: Have the agent run 10+ parallel Bash tool calls. Observe that the batch fails with hook validation errors even though individual calls succeed. + +## Actual Result + +Issue 1: Agent hangs waiting for permission response that the user has no way to grant. +Issue 2: All parallel calls fail with "Sibling tool call errored" cascade. + +## Expected Result + +Issue 1: Permission dialog should reliably appear whenever the agent requests tool approval. +Issue 2: Parallel tool calls should either all be validated independently, or failures should be isolated rather than cascading. + +## Acceptance Criteria + +- [ ] Permission request dialog reliably appears in the web UI when the agent needs tool approval +- [ ] Parallel Bash tool calls do not cascade-fail due to hook/permission validation errors +- [ ] Root cause identified for each sub-issue (web UI, Claude Code SDK, or hook system)